The hero inside the smoke: who covers and who attacks

The first two chapters of this study measured the play: how many lanes were pushed, where the group went, where the opposing carry was. None of them looked at who was under the smoke. This one looks, and the question that opened it is very specific.

The theory to test: there are attack heroes in smoke (they enter to start the fight) and cover heroes (they exist so that the fight that goes wrong doesn't cost the match). The Winter Wyvern would be of the second type; the Lion, of the first; Mirana and Dark Willow would do both.

This chapter is the only one in the study that runs on a larger base than TI: there are 2,628 smokes from 240 maps, adding up the entire The International 2026 and the Esports World Cup 2026. The reason is arithmetic: with a single tournament, half of the heroes got a dozen appearances, and the rate on top of ten moves is not measured. The base change cost two first-round completions, and they are recorded at the bottom of the page.

To answer this, each hero is measured on both sides:

• With him under the offensive smoke: the one that ends in the opponent's field. If he is attacking, the conversion goes up when he is in the smoke.
• Against him: he is on the roster of the team that *suffers* the smoke. If he is a cover, the opponent's smoke yields less.

And the two measures also appear discounting the gold advantage, in bands of 5,000, because without it "good hero in smoke" measures to a large extent "the team that picked this hero was winning".

The four of the question

The four heroes of the question

Offensive smoke (the one that ends in the opponent's field) with the hero under the smoke, and the opponent's smoke against his team

Hero	With him underneath	Converts	explode	liquid slaughter.	Go up in smoke	Discounted conversion	Smokes suffered	Opponent converts	discounted
Winter Wyvern	134	51%	27%	+1,0	60% of 225	+0,2 pp	366	39%	−3,1 pp
Lion	28	54%	29%	+0,5	45% of 62	+5,2 pp	88	35%	−5,8 pp
Mirana	163	49%	28%	+0,7	60% of 272	−1,5 pp	452	40%	−2,5 pp
Dark Willow	46	41%	28%	+0,9	26% of 175	−6,2 pp	401	37%	−5,1 pp

Go up in smoke: Of his team's offensive smokes, how many he actually entered: a hero who is left out cannot be judged by what the smoke yielded · discounted: the difference in percentage points within equal ranges of gold advantage. In the attack column, up is better; in the cover column, down.

The Winter Wyvern: half of the theory survives, and it's the half that matters. It's under 134 offensive smokes, which converted 51% and exploded 27%. Discounting the score, what it adds to the offense is +0.2 conversion point: zero, with uncomfortable accuracy. On the other hand, smokar against a team that has Wyvern converts 3.1 points less (366 shots conceded).

So: she's not an attack hero in smoke: that's measured and it's the firm find. But her cover is modest, it's not the big effect that the first round showed (it was −8.2 points with TI's 96 maps; dropped to −3.1 with 240). And notice that she enters 60% of the team's offensive smokes: she goes with it, and doesn't change the offense.

The Lion became the opposite of what the first round said. With 240 maps, he adds +5.2 conversion points when he is under the smoke (now the initiator that fame promises) and the opponent's smoke against his team converts 5.8 points less. In other words: in the large sample, he is both. What made the reading turn was n: there were 23 attack shots and 55 conceded on TI; now it's 28 and 88, and the cover signal has changed from +5.3 to −5.8. A hero with Hex and two-second stun doing both is, honestly, the more plausible outcome of the two.

Mirana is the stable line of the study. −1.5 conversion with it down, −2.5 conversion conceded, in 163 and 452 moves. It draws on both sides in the two rounds, which is what "does both" means when what is measured is effect, not sensation.

The Dark Willow switched sides on offense. With her in the smoke, the smoke yields −6.2 points (in the first round it was +1.8), and the cover was −5.1. Add to that the fact that she only gets into 26% of the team's offensive smokes (the lowest rate of the four) and the picture is coherent: she's good for the team and bad for the smoke. What she does (Cursed Crown planted, Bramble Maze cutting the lane) works when she's behind, not when she's inside the siege.

Who really makes the offensive smoke pay off

Who makes the offensive smoke pay off

ordered by conversion discounted the gold advantage · minimum of 30 smokes with him underneath

Hero	With him underneath	Converts	explode	liquid slaughter.	Go up in smoke	Discounted conversion	Smokes suffered	Opponent converts	discounted
Keeper Of The Light	42	64%	24%	+0,9	39% of 107	+15,4 pp	265	40%	−1,2 pp
Shadow Demon	38	58%	18%	+0,6	75% of 51	+15,4 pp	66	56%	+19,1 pp
Kez	47	60%	17%	+1,5	36% of 132	+10,5 pp	198	34%	−6,8 pp
Enchantress	31	61%	19%	+0,9	41% of 76	+10,5 pp	129	39%	−5,1 pp
Rubick	65	60%	18%	+1,4	47% of 139	+9,8 pp	217	43%	+2,3 pp
Lina	53	58%	23%	+0,8	34% of 157	+8,8 pp	320	39%	−2,3 pp
Earth Spirit	78	64%	17%	+1,4	30% of 257	+8,3 pp	425	40%	−3,0 pp
Clockwerk	99	56%	27%	+0,4	42% of 235	+8,0 pp	402	45%	+2,9 pp
Pangolier	52	56%	23%	+1,1	47% of 111	+5,7 pp	168	43%	+3,7 pp
Centaur	64	59%	16%	+1,5	27% of 233	+5,3 pp	467	39%	−4,1 pp
Jakiro	32	50%	25%	+0,4	45% of 71	+5,1 pp	117	45%	+4,7 pp
Snapfire	34	56%	21%	+1,0	29% of 118	+4,9 pp	219	43%	+0,8 pp
Slardar	44	59%	16%	+1,4	39% of 114	+3,7 pp	166	34%	+0,6 pp
Hoodwink	209	55%	28%	+0,9	51% of 412	+3,2 pp	615	44%	+4,5 pp
Ember Spirit	65	55%	23%	+0,9	24% of 274	+3,1 pp	384	35%	−3,9 pp
Bane	34	62%	18%	+1,5	27% of 128	+3,0 pp	236	45%	+4,8 pp
Undying	100	52%	25%	+0,9	48% of 209	+1,1 pp	360	39%	−2,0 pp
Windranger	88	50%	23%	+0,7	39% of 224	+0,5 pp	357	40%	−3,4 pp
Tiny	79	54%	22%	+0,7	54% of 145	+0,3 pp	239	39%	+1,8 pp
Winter Wyvern	134	51%	27%	+1,0	60% of 225	+0,2 pp	366	39%	−3,1 pp
Doom	42	62%	21%	+1,0	29% of 143	−0,6 pp	218	36%	−1,1 pp
Ringmaster	50	44%	34%	+0,4	33% of 151	−0,8 pp	232	43%	±0,0 pp
Mirana	163	49%	28%	+0,7	60% of 272	−1,5 pp	452	40%	−2,5 pp
Shadow Fiend	47	53%	17%	+1,3	28% of 170	−2,2 pp	257	33%	−1,7 pp
Alchemist	30	47%	37%	+0,2	48% of 62	−2,2 pp	97	48%	+11,8 pp
Tusk	83	42%	36%	+0,2	66% of 125	−3,9 pp	206	47%	+2,3 pp
Dark Willow	46	41%	28%	+0,9	26% of 175	−6,2 pp	401	37%	−5,1 pp
Necrophos	42	43%	21%	+0,4	26% of 162	−6,4 pp	267	36%	−2,3 pp
Dark Seer	52	42%	38%	+0,1	43% of 120	−7,6 pp	198	45%	+2,4 pp

Go up in smoke: Of his team's offensive smokes, how many he actually entered: a hero who is left out cannot be judged by what the smoke yielded · discounted: the difference in percentage points within equal ranges of gold advantage. In the attack column, up is better; in the cover column, down.

The raid list isn't made of damage heroes: it's made of control that closes the siege. Keeper of the Light (+15.4 conversion, −8.5 backfire), Shadow Demon (+15.4 / −11.2), Kez (+10.5 / −11.9), Enchantress (+10.5), Rubick (+9.8 / −9.2), Lina (+8.8), Earth Spirit (+8.3 / −10.5). Everyone turns "we've come close" into "the enemy can't get out": by imprisonment, by pure disable, or by slowness.

And the end of the list is as informative as the beginning: Dark Seer (−7.6 conversion and +10.9 backfire), Necrophos (−6.4), and Dark Willow (−6.2). The Dark Seer is the didactic case: Vacuum is an initiation that depends on hitting an already positioned group, and smoke that misses the first ability becomes the worst type of teamfight: five heroes on the opponent's field without any surprise.

A reading caution: the column "enters the smoke" changes the direction of the line. The Shadow Demon enters 75% of the team's smokes and improves the attack by 15 points: he is made for that. The Keeper of the Light enters 39% and improves as much as: he chooses the time. The Mirana enters 60% (163 moves, one of the highest n on the list) and is worth −1.5: present and neutral.

Who makes the opponent's smoke yield less

Who makes the opponent's smoke yield less

ordered by conversion suffered, minus gold advantage · minimum of 40 smokes suffered with him in the squad

Hero	With him underneath	Converts	explode	liquid slaughter.	Go up in smoke	Discounted conversion	Smokes suffered	Opponent converts	discounted
Treant Protector	17	88%	6%	+2,0	24% of 72	+37,1 pp	101	18%	−23,0 pp
Shadow Shaman	7	57%	0%	+1,9	35% of 20	+5,7 pp	49	35%	−12,8 pp
Phantom Lancer	15	73%	20%	+1,5	39% of 38	+20,2 pp	68	35%	−10,2 pp
Kez	47	60%	17%	+1,5	36% of 132	+10,5 pp	198	34%	−6,8 pp
Lycan	11	36%	18%	+0,7	15% of 71	−20,4 pp	95	28%	−6,4 pp
Enigma	21	43%	33%	+1,0	28% of 76	−18,6 pp	121	39%	−5,8 pp
Lion	28	54%	29%	+0,5	45% of 62	+5,2 pp	88	35%	−5,8 pp
Nature's Prophet	18	50%	28%	+0,6	16% of 112	−20,0 pp	164	34%	−5,6 pp
Dark Willow	46	41%	28%	+0,9	26% of 175	−6,2 pp	401	37%	−5,1 pp
Enchantress	31	61%	19%	+0,9	41% of 76	+10,5 pp	129	39%	−5,1 pp
Centaur	64	59%	16%	+1,5	27% of 233	+5,3 pp	467	39%	−4,1 pp
Ember Spirit	65	55%	23%	+0,9	24% of 274	+3,1 pp	384	35%	−3,9 pp

Go up in smoke: Of his team's offensive smokes, how many he actually entered: a hero who is left out cannot be judged by what the smoke yielded · discounted: the difference in percentage points within equal ranges of gold advantage. In the attack column, up is better; in the cover column, down.

Here's the other side of the same coin, and it's the table a draft should look at before the fifth pick.Treant Protector is the biggest measured cover, and by a margin that is not discussed: the opponent's smoke converts 23 points less and explodes 44% of the time, in 101 shots conceded. Living Armor from a distance and Overgrowth on top of the siege undo the play without him needing to be there: it's the cover in its pure state.

After it come Shadow Shaman (−12.8), Phantom Lancer (−10.2), Kez (−6.8), Lycan (−6.4), and Enigma (−5.8). They are different mechanisms with the same effect: long disable that reverses the fight, illusions that make the siege miss the target, or an area response that switches hand initiation.

And the bottom of the table is a draft warning in the opposite direction: Shadow Demon (+19.1), Alchemist (+11.8) and Luna (+11.4) make the opponent's smoke yield MORE. The Shadow Demon appears on both ends (it's the biggest offense and the biggest "anti-cover" on the list) and that's no contradiction: a hero that exists to amplify a focus amplifies the focus of whoever gets there first.

And my hero?

The two tables above cut to 30 appearances, because below that the rate does not support a ranking. But niche heroes are also picked, and the question "what about mine?" deserves an answer. The full table has the 90 heroes that appeared in a non-anecdotal way on the 240 maps, ordered by presence in smoke, with the n of each line in view, and the line dimmed when there are less than 15 appearances, which is where the number indicates direction and not measure.

All 90 Heroes · Open the table

Ordered by presence in smoke: the two tables above already order by effect · dimmed line = less than 15 appearances, direction and not measured

Hero	With him underneath	Converts	explode	liquid slaughter.	Go up in smoke	Discounted conversion	Smokes suffered	Opponent converts	discounted
Hoodwink	209	55%	28%	+0,9	51% of 412	+3,2 pp	615	44%	+4,5 pp
Mirana	163	49%	28%	+0,7	60% of 272	−1,5 pp	452	40%	−2,5 pp
Winter Wyvern	134	51%	27%	+1,0	60% of 225	+0,2 pp	366	39%	−3,1 pp
Undying	100	52%	25%	+0,9	48% of 209	+1,1 pp	360	39%	−2,0 pp
Clockwerk	99	56%	27%	+0,4	42% of 235	+8,0 pp	402	45%	+2,9 pp
Windranger	88	50%	23%	+0,7	39% of 224	+0,5 pp	357	40%	−3,4 pp
Tusk	83	42%	36%	+0,2	66% of 125	−3,9 pp	206	47%	+2,3 pp
Tiny	79	54%	22%	+0,7	54% of 145	+0,3 pp	239	39%	+1,8 pp
Earth Spirit	78	64%	17%	+1,4	30% of 257	+8,3 pp	425	40%	−3,0 pp
Ember Spirit	65	55%	23%	+0,9	24% of 274	+3,1 pp	384	35%	−3,9 pp
Rubick	65	60%	18%	+1,4	47% of 139	+9,8 pp	217	43%	+2,3 pp
Centaur	64	59%	16%	+1,5	27% of 233	+5,3 pp	467	39%	−4,1 pp
Lina	53	58%	23%	+0,8	34% of 157	+8,8 pp	320	39%	−2,3 pp
Dark Seer	52	42%	38%	+0,1	43% of 120	−7,6 pp	198	45%	+2,4 pp
Pangolier	52	56%	23%	+1,1	47% of 111	+5,7 pp	168	43%	+3,7 pp
Ringmaster	50	44%	34%	+0,4	33% of 151	−0,8 pp	232	43%	±0,0 pp
Shadow Fiend	47	53%	17%	+1,3	28% of 170	−2,2 pp	257	33%	−1,7 pp
Kez	47	60%	17%	+1,5	36% of 132	+10,5 pp	198	34%	−6,8 pp
Dark Willow	46	41%	28%	+0,9	26% of 175	−6,2 pp	401	37%	−5,1 pp
Slardar	44	59%	16%	+1,4	39% of 114	+3,7 pp	166	34%	+0,6 pp
Necrophos	42	43%	21%	+0,4	26% of 162	−6,4 pp	267	36%	−2,3 pp
Keeper Of The Light	42	64%	24%	+0,9	39% of 107	+15,4 pp	265	40%	−1,2 pp
Doom	42	62%	21%	+1,0	29% of 143	−0,6 pp	218	36%	−1,1 pp
Shadow Demon	38	58%	18%	+0,6	75% of 51	+15,4 pp	66	56%	+19,1 pp
Bane	34	62%	18%	+1,5	27% of 128	+3,0 pp	236	45%	+4,8 pp
Snapfire	34	56%	21%	+1,0	29% of 118	+4,9 pp	219	43%	+0,8 pp
Jakiro	32	50%	25%	+0,4	45% of 71	+5,1 pp	117	45%	+4,7 pp
Enchantress	31	61%	19%	+0,9	41% of 76	+10,5 pp	129	39%	−5,1 pp
Alchemist	30	47%	37%	+0,2	48% of 62	−2,2 pp	97	48%	+11,8 pp
Largo	29	45%	38%	+0,3	21% of 139	−1,2 pp	228	44%	−0,2 pp
Drow Ranger	29	45%	38%	+0,3	26% of 113	−1,9 pp	224	46%	−2,4 pp
Life Stealer	28	46%	32%	+0,7	42% of 67	+0,1 pp	109	48%	+3,2 pp
Lion	28	54%	29%	+0,5	45% of 62	+5,2 pp	88	35%	−5,8 pp
Axe	26	38%	35%	+0,4	23% of 113	−10,7 pp	215	50%	+8,2 pp
Io	26	54%	38%	+0,3	45% of 58	+9,7 pp	105	47%	+1,4 pp
Clinkz	25	52%	28%	+0,4	28% of 89	+6,3 pp	132	40%	−3,0 pp
Sven	25	48%	40%	+0,2	44% of 57	−9,8 pp	100	43%	+2,1 pp
Techies	23	48%	35%	+0,4	58% of 40	+9,2 pp	59	37%	−3,9 pp
Beastmaster	22	45%	27%	+0,3	31% of 72	+2,1 pp	159	45%	+0,8 pp
Elder Titan	22	45%	36%	+0,1	76% of 29	−2,3 pp	78	46%	−3,9 pp
Enigma	21	43%	33%	+1,0	28% of 76	−18,6 pp	121	39%	−5,8 pp
Invoker	20	40%	45%	+0,5	17% of 117	−12,1 pp	208	47%	+7,0 pp
Terrorblade	20	50%	30%	−0,1	28% of 71	−2,3 pp	112	47%	+6,8 pp
Nature's Prophet	18	50%	28%	+0,6	16% of 112	−20,0 pp	164	34%	−5,6 pp
Earthshaker	18	61%	33%	+1,6	45% of 40	+7,8 pp	61	41%	+1,8 pp
Puck	17	41%	29%	−0,2	16% of 109	+2,7 pp	229	53%	+4,8 pp
Treant Protector	17	88%	6%	+2,0	24% of 72	+37,1 pp	101	18%	−23,0 pp
Phantom Lancer	15	73%	20%	+1,5	39% of 38	+20,2 pp	68	35%	−10,2 pp
Witch Doctor	14	43%	29%	+0,3	41% of 34	−7,3 pp	79	46%	+1,7 pp
Viper	13	62%	23%	+0,7	24% of 55	+11,4 pp	92	48%	+2,8 pp
Underlord	12	50%	33%	+0,3	12% of 98	−8,3 pp	193	44%	−1,2 pp
Timbersaw	12	50%	42%	+0,7	16% of 73	−5,7 pp	137	49%	+6,0 pp
Monkey King	12	67%	25%	+1,1	22% of 54	+16,8 pp	89	43%	+0,1 pp
Phoenix	12	67%	25%	+0,4	48% of 25	+22,1 pp	62	42%	−3,8 pp
Sand King	12	25%	58%	−0,8	35% of 34	−22,3 pp	56	41%	−3,2 pp
Storm Spirit	11	73%	9%	+1,6	18% of 62	+38,3 pp	117	49%	+4,5 pp
Lycan	11	36%	18%	+0,7	15% of 71	−20,4 pp	95	28%	−6,4 pp
Bounty Hunter	11	55%	27%	+0,7	21% of 52	+0,2 pp	66	30%	−3,5 pp
Batrider	11	55%	18%	+1,3	46% of 24	+14,0 pp	36	42%	−8,9 pp
Magnus	11	36%	27%	+0,2	73% of 15	−7,3 pp	27	41%	+0,2 pp
Dragon Knight	10	40%	20%	+1,2	40% of 25	−10,5 pp	66	50%	+8,5 pp
Gyrocopter	8	25%	50%	−0,1	36% of 22	−11,7 pp	37	51%	+13,1 pp
Void Spirit	8	50%	38%	+0,9	44% of 18	−11,5 pp	36	53%	+8,2 pp
Shadow Shaman	7	57%	0%	+1,9	35% of 20	+5,7 pp	49	35%	−12,8 pp
Huskar	7	14%	29%	−0,9	30% of 23	−25,9 pp	45	44%	+9,3 pp
Luna	7	43%	57%	−1,1	78% of 9	+8,6 pp	44	59%	+11,4 pp
Dazzle	7	57%	14%	+0,1	32% of 22	+10,8 pp	28	32%	−12,4 pp
Lich	6	33%	17%	+0,7	15% of 40	−15,3 pp	70	47%	+1,7 pp
Queen of Pain	6	33%	50%	−0,5	46% of 13	−2,6 pp	28	57%	+14,9 pp
Sniper	6	33%	33%	0,0	60% of 10	−3,1 pp	19	53%	+0,4 pp
Tidehunter	5	60%	20%	+1,4	14% of 37	−3,1 pp	71	41%	−3,2 pp
Grimstroke	5	80%	0%	+2,0	17% of 30	+8,7 pp	59	54%	+4,5 pp
Night Stalker	5	80%	20%	+2,4	24% of 21	+32,3 pp	42	60%	+7,8 pp
Faceless Void	5	40%	40%	+1,0	56% of 9	−8,8 pp	22	77%	+25,1 pp
Ogre Magi	5	60%	20%	+1,0	71% of 7	−13,9 pp	17	29%	−4,3 pp
Muerta	5	100%	0%	+3,8	42% of 12	+46,4 pp	16	44%	+1,6 pp
Lone Druid	4	25%	25%	+0,5	5% of 80	−20,1 pp	179	42%	+1,9 pp
Spectre	4	75%	25%	+1,5	9% of 44	+38,4 pp	67	40%	−3,7 pp
Naga Siren	4	0%	100%	−1,7	40% of 10	−46,1 pp	27	37%	−2,6 pp
Pudge	4	75%	0%	+2,8	36% of 11	+17,0 pp	24	33%	−9,6 pp
Zeus	4	50%	0%	+1,8	67% of 6	+4,4 pp	21	38%	−1,9 pp
Disruptor	3	67%	0%	+1,3	18% of 17	−4,6 pp	36	58%	+12,1 pp
Crystal Maiden	3	67%	33%	−0,7	43% of 7	+38,7 pp	34	50%	+10,5 pp
Spirit Breaker	3	33%	67%	−1,7	25% of 12	+7,7 pp	31	74%	+12,3 pp
Slark	3	100%	0%	+3,0	38% of 8	+28,0 pp	24	17%	−9,0 pp
Oracle	3	100%	0%	+2,3	43% of 7	+42,3 pp	20	50%	+24,2 pp
Templar Assassin	2	100%	0%	+2,5	17% of 12	+46,0 pp	37	78%	+18,5 pp
Razor	2	50%	0%	+2,5	13% of 16	−11,9 pp	22	41%	−3,1 pp
Dawnbreaker	1	0%	0%	+1,0	2% of 59	−51,0 pp	78	37%	+5,4 pp
Kunkka	0	0%	0%	0,0	0% of 17	—	23	26%	−12,6 pp

Two warnings to read, because a long list invites error. First: don't read the top of the effect as ranking. Muerta adds +46 conversion points in 5 moves and the Templar Assassin +46 in 2: that's noise, not talent, and that's exactly why the table is sorted by presence and not by effect. Second: the two columns of the end (cover) have a much larger sample than the middle (attack) columns, because a hero is an opponent in four times more moves than he is an ally under the smoke. A line can be thin on one side and solid on the other: the Kunkka appears with zero smokes at the bottom and 23 suffered.

The third wave: who plays away from the team and connects later

There's a family of heroes that the attack table misjudges by construction: the ones that shouldn't be in smoke. It's what Astini calls the *third wave*, and it's not about the literal wave, it's the name of the heroes who can play the furthest away from the team and get into the fight after it's started. Three different mechanics do this:

• portal: Underlord, Nature's Prophet, Io, Keeper of the Light, Dawnbreaker, Spectre: they (or the team) traverse the map from above;
• global effect: Invoker with Sunstrike, Centaur with Stampede: participate without leaving the place;
• Mobility: Ember Spirit, Puck, Pangolier, Storm Spirit, Monkey King, Void Spirit, Slark: traverse the map faster than everyone else.

To measure this, "connect" needed a ruler in the replay: the hero was more than 2,500 from the group's destination in the second of the squeeze and came within 1,800 of him within 90 seconds. It's the closest the die gets to "he showed up in the fight".

The first column already answers the question. Portal heroes go up in smoke in 22% of the team's smokes, against 41% of all other heroes: just over half. And it's not a sample accident: they're 7,220 units away from the destination when the smoke leaves (the median of the rest is 5,723), that is, on the other side of the map. They don't participate in the gesture of going along.

The last part of the table is what closes the argument, and with the enlarged base it became more favorable to the thesis. With the portal hero inside the smoke, the bid converts 43% and explodes 34%. With him out, but connecting, it converts 51% and explodes 26%: better than with him inside. With him out and not connecting, it drops to 34% / 37%.

In other words: for these heroes, staying out costs nothing, and can yield more, as long as they arrive. And while the rest of the team traverses the map smoked, they make 641 creep gold in the 90 seconds of the play, against 395 in the rest of the tournament. The smoke yields more and the hero yields more.

The mobility family connects more and faster than all (72% of the time, in 26 seconds) and for it inside and outside they tie (44% against 43%). The practical reading is the same: a Ember Spirit under the smoke is spending the skill that defines him to do what any hero would do.

The one with a global effect is the only one in which "not connecting" does not sink the bid: 43% conversion without arriving, equal to 43% when it arrives. It makes sense by the mechanism: Sunstrike and Stampede deliver the effect from wherever they are.

The heroes who go up in smoke the least

"connect" is measured in the replay: it was more than 2.500 of fate in the second of the squeeze and reached 1.800 His way up to 90s

Hero	Go up in smoke	stayed away	connected	How soon	Distance in Grip	Creep gold in the 90s	With him inside	Outside, connecting	Outside, without connecting
Dawnbreaker	4%	70	53%	30s	7.447	685	0% / 33%	49% / 35%	27% / 36%
Spectre	10%	58	62%	37s	8.640	647	71% / 29%	50% / 31%	32% / 41%
Underlord	13%	133	53%	34s	7.816	589	55% / 27%	51% / 24%	32% / 43%
Lich	16%	45	84%	23s	5.104	123	18% / 45%	53% / 26%	43% / 43%
Nature's Prophet	17%	145	50%	32s	7.420	1.008	38% / 32%	60% / 26%	38% / 29%
Lone Druid	19%	116	22%	34s	7.939	868	53% / 27%	38% / 35%	44% / 36%
Tidehunter	19%	44	73%	33s	5.942	451	25% / 50%	53% / 34%	33% / 42%
Timbersaw	20%	88	69%	28s	5.982	693	42% / 46%	48% / 26%	26% / 44%
Invoker	20%	137	67%	30s	6.137	650	33% / 49%	47% / 25%	40% / 40%
Puck	22%	133	71%	28s	5.571	566	35% / 44%	29% / 49%	15% / 46%
Monkey King	22%	59	61%	31s	6.097	540	47% / 21%	42% / 36%	39% / 26%
Bounty Hunter	23%	46	72%	18s	4.661	52	53% / 29%	52% / 30%	23% / 31%
Viper	23%	61	67%	30s	5.379	590	35% / 39%	32% / 49%	20% / 40%
Lycan	23%	71	59%	29s	6.146	587	33% / 33%	64% / 19%	52% / 17%
Ember Spirit	24%	257	77%	26s	5.182	526	46% / 31%	46% / 25%	27% / 33%
Beastmaster	24%	96	72%	28s	5.799	527	34% / 34%	33% / 33%	26% / 30%

the last three columns are the outcome of the THROW (converted / exploded) with the hero in each situation · the hero line counts one record per smoke of his team, so the n is for the hero× bid, not for the bid

Hero for hero, the extremes are didactic. Dawnbreaker enters 4% of her team's smokes: she is the definition of a third wave, because Solar Guidance arrives anywhere on the map and doesn't need smoke to do so. Spectre 10%, Underlord 13%, Lich 16%, Nature's Prophet 17%, Lone Druid 19%.

Notice that the list has names that are not in any of the three families: Lich, Lone Druid, Clinkz, Necrophos. They have no global connection: they play far away for another reason (pushing another lane, invisibility, accelerated farming). This is an honest reminder: "third wave" describes a behavior, and the behavior appears in heroes that the theory did not predict.

And the counterexample that keeps the reading standing: when the portal hero doesn't connect, the throw is the worst cut on this page (34% / 37%). What the table says is not "ignore a smoke". It is: *if your hero throws far, he doesn't need to go into the smoke, but he needs to get in.*

What to do about it

1. Cover is safe, it's not attack. A Treant, a Shadow Shaman or a Wyvern on your team takes 3 to 23 points from the opponent's smoke conversion, and doesn't add anything to yours. You still need control that closes in on the smoke.
2. If your initiator relies on hitting a target (Dark Seer), the smoke is riskier, not less. Discounting the score, it adds up to 10.9 backfire points. With it, the smoke has to have an objective, and the general guide shows what the objective in the window does to the account.
3. If the opponent has Shadow Demon, Alchemist or Luna, the smoke is worth more. It is the opposite of item 1: they are the squads in which the hunt yields 11 to 19 points more.
4. A hero who doesn't go into smoke can't be charged for what the smoke yielded. The "go into smoke" column exists for that: Dark Willow goes into 31% of her team's smokes: her effect isn't on the attack, it's on the map.
5. If your team has a third-wave hero, count on him out of the smoke and cover the connection. Portal and mobility yield the same (or more) connecting later than going along, and still bring 700 to 1,000 creep gold in the window. What can't happen is the play without him: without connecting, the conversion drops from 47% to 30%.

What the larger sample changed

The first round of this page came out with the 96 maps of TI15: 992 moves. With the EWC inside, the base became 2,628, and two conclusions changed owners. They are recorded, because a study that only publishes what survives does not let anyone learn:

• The Lion turned the other way around. With 96 maps, the opponent's smoke yielded more against his team (+5.3 points) and he didn't add anything in attack. With 240, he adds +5.2 conversion in smoke and the opponent converts 5.8 less against him: he went from "only good facing forward" to "does both". What changed was n (55 shots conceded against 88) and it is exactly the sample size in which a five-point signal becomes noise.
• The Dark Willow switched sides in attack, from +1.8 to −6.2. The part that survived was the one that was already the most interesting: she almost doesn't go into smoke (31% in the first round, 26% now) and covers from the outside.
• The Winter Wyvern 's cover shrank from −8.2 to −3.1. The conclusion that stood is the other half, and she didn't move in either of the two rounds: she doesn't add attack (+3.5 before, +0.2 now).
• The third wave became stronger. Portal connecting went from "equal to being inside" to better than being inside (51%/26% against 43%/34%), and the difference in presence in the smoke (22% against 41%) remained.

What hasn't changed in any of the rounds: Mirana is neutral on both sides, control that closes siege is what makes offensive smoke pay, and initiation that depends on hitting a specific target makes everything worse.

Method, and where it has a limit

• Base. 2,628 game smokes from 240 maps: the entire The International 2026 (139 maps) plus the Esports World Cup 2026 (101). It is the only page in the study that leaves the TI clip: the other two remain on 96 maps with the six teams closely monitored. The attack cut uses the 1,520 offensive ones (the ones that end up in the opponent's field 45 seconds later); the cover one uses them all, because suffering smoke does not depend on the smoke being offensive.
• Two tournaments, two patches. The EWC was in July and the TI in August. A hero who changed strength between the two falls into the same line, and the dice don't separate that.
• Whoever was under the smoke exits the replay: companion alive within the 1,200 radius of the player who pressed it, at the exact second of use. Dead hero does not enter, even if glued to the radius.
• The ruler is the same as the rest of the study: converted and backfire measured in the 120-second window.
• Every cut has the discounted version the gold advantage, in bands of 5,000, with a weighted average of the differences.
• One move counts for multiple heroes. A smoke with three heroes underneath falls into the line of three: that's what the question asks, and that's also why the lines don't add up to 549.
• Where the number is thin. The highlight tables cut into 30 appearances (attack) and 40 suffered (cover). The complete list goes down to 5 appearances to answer "what about my hero?", and in it the line with less than 15 is dimmed: 42 of the 90 heroes are in this range. The n is in every line, and the order is by presence precisely so that a thin sample does not lead the table.
• "Connect" is a ruler, not an intent reading. The hero enters the account as "connected" if he was more than 2,500 from the destination in the squeeze and reached less than 1,800 from him in 90 seconds. This does not distinguish who went to help from who passed there, and the line of the third wave counts one record per smoke of the team, so its n is for the hero× bid.
• What this study doesn't see: items, level, and skill order. A Wyvern with Aghanim's at 25 minutes and one without, at 15, fall in the same line. And composition is not a cause: a team that picked Prophet plays the map in a way that the hero only partially explains.